We conducted a prospective cohort study at the Department of Cardiology, Zhongshan Hospital of Fudan University, Shanghai City, China, from September 1, 2015, to December 31, 2016. Patients with HFrEF were prospectively evaluated for inclusion in this study. HFrEF was diagnosed according to the current consensus statements of the American Heart Association [1] and the 2018 Chinese guidelines for the diagnosis and treatment of heart failure [14]. The inclusion criteria were as follows: (1) symptoms or signs of HF, (2) N-terminal prohormone of brain natriuretic peptide (NT-proBNP) > 125 ng/L; (3) left ventricular ejection fraction (LVEF) < 40%; and (4) New York Heart Association (NYHA) functional class ≥ II. The exclusion criteria were: (1) congenital heart disease, (2) acute coronary syndrome in the last 30 days, (3) pericardial disease, (4) pacemaker or other conditions precluding patients from CMRI, (5) severe anemia (hemoglobin < 7 g/dL), (6) chronic obstructive pulmonary disease, GOLD stage 3 or 4, and (7) estimated glomerular filtration rate < 30 mL/min/1.73 m². The study protocol conformed to the Declaration of Helsinki, and its subsequent amendments and was approved by the local ethics committee of Zhongshan Hospital, Fudan University. All participants provided written informed consent.

Covariates in the present study included general information, demographics, variables that could affect the ratio of sST2/LVMI or cardiac mortality, and HF hospitalization, based on our clinical experiences and previous reports.

Demographic data and clinical variables, including age, sex, body mass index, diastolic blood pressure, systolic blood pressure, heart rate, NYHA functional class, medical history, and cardiovascular risk factors (smoking, hypertension, and diabetes mellitus), were collected. Fasting venous blood was collected within 12 h after admission. After centrifugation at 3000 rpm for 15 min, the plasma was extracted and stored at 80 °C. Biochemical variables, including hematocrit, hemoglobin, white blood cells, NT-proBNP, sodium, creatinine, blood urea nitrogen, serum uric acid, albumin, total bilirubin, total cholesterol, high-density lipoprotein cholesterol, and hypersensitive C-reactive protein were measured. Serum biomarkers of myocardial fibrosis (sST2, procollagen III amino terminal propeptide [PIIINP], procollagen type I carboxy-terminal propeptide [PICP]) were assayed simultaneously using the respective ELISA kits. The characteristics of the assays were as follows: sST2 ELISA (Critical Diagnostics, San Diego, CA, USA, Catalog No. BC-1065E): average intra-assay coefficient of variation (CV) of 5.1%, detection limit of 1.8 ng/mL; PIIINP ELISA (MyBioSource, San Diego, CA, USA; Catalog No. MBS703383): intra-and inter-assay CV of less than 10%, detection range 0.125–8 ng/mL; PICP ELISA (Elabscience, Wuhan, China; Catalog No. E-EL-H6030): intra- and inter-assay CV less than 10%, detection range 0.78 to 50 ng/mL. Serum levels of PIIINP were measured using the Roche Elecsys autoanalyzer (Cobas e602), with intra-assay CV of 1.2%–4.1%, inter-assay CV of 3.75%, and a detection range of 5 to 1200 ng/mL.

Echocardiography was performed according to the American Society of Echocardiography guidelines [15]. All participants underwent transthoracic echocardiography by board-certified physicians using a Philips iE33 ultrasound machine (Philips Medical Systems, Eindhoven, The Netherlands) equipped with S5–1 and X3–1 probes. Left atrial diameter, LVEF, left ventricular end-diastolic diameter, and interventricular septal thickness were analyzed.

As described in our previous work [16], all subjects underwent clinical CMRI scans performed by two dedicated CMRI technologists in a 1.5-T CMRI system (MAG-NETOM Area, Siemens Healthcare, Erlangen, Germany) with an 18-channel phased-array cardiovascular coil. CMRI data analysis was performed using the dedicated software Argus (Siemens Medical Solution, Erlangen, Germany) by an observer blinded to all clinical data. Left ventricular mass (LVM) was determined by tracing the epicardial and endocardial borders of each slice at end-diastole, summing the myocardial volume of all slices, and multiplying by myocardial density (1.05 g/mL) [17]. LVM was indexed to body surface area (LVMI). Other CMRI variables were measured using methods previously described [16].

Patients were followed up by telephone calls and ambulatory visits at 9-month intervals. The primary outcome was a combined endpoint consisting of HF rehospitalization and cardiovascular death. The follow-up time was calculated from the time of discharge to the primary outcome, or 9 months after discharge. Endpoints were assessed by all coauthors.

Data were expressed as mean (standard deviation) for Gaussian distribution or median (min, max) for skewed distribution of continuous variables and as numbers and percentages for categorical variables. The χ² test (categorical variables), one-way ANOVA test (normal distribution), or Kruskal–Wallis H test (skewed distribution) was used to detect the differences among patients with different sST2/LVMI ratios (tertiles). We used univariate and multivariate Cox proportional hazards regression models to test the link between the sST2/LVMI ratio and the primary outcome with three distinct models. Model 1 was an unadjusted model. Model 2 was a minimally adjusted model only for sociodemographic variables. Model 3 was a fully adjusted model. Because Cox proportional hazards regression model-based methods are often considered inadequate to address nonlinear relationships, nonlinearity between the sST2/LVMI ratio and the primary outcome was addressed using a Cox proportional hazards regression model with cubic spline functions and smooth curve fitting (penalized spline method). If nonlinearity was detected, we first calculated the inflection point using the recursive algorithm and then constructed a two-piecewise Cox proportional hazards regression model on both sides of the inflection point. Subgroup analyses were performed using a stratified Cox proportional hazards regression model. For each continuous variable, we first converted it to a categorical variable according to the clinical cut point or tertile and then performed an interaction test. Tests for effect modification of subgroup indicators were followed by the likelihood ratio test. Log-rank tests for Kaplan–Meier survival curves were performed to test the prognostic value of various sST2/LVMI ratios.

Data were analyzed using the statistical software packages R (http://​www.​R-project.​org, The R Foundation) and EmpowerStats (http://​www.​empowerstats.​com, X&Y Solutions, Inc, Boston, MA). All statistical tests were two-sided, and a P-value < 0.05 was considered statistically significant.

After a baseline evaluation, 45 patients were enrolled. After 9 months of follow-up, 15 patients had reached the primary endpoint (33.3%), of whom two patients had died and 13 had been rehospitalized due to worsening HF. No patient was lost to follow-up. We show the baseline characteristics of the selected participants in Table 1, according to the tertile of the sST2/LVMI ratio. The average age was 48 ± 14 years, and approximately 20% were women. Patients with the highest sST2/LVMI ratio (Q3) had significantly higher blood sST2 levels, and they were more likely to have been prescribed angiotensin converting enzyme inhibitors or angiotensin receptor blockers than other groups. Opposite patterns were observed for the myocardium post-contrast T1 time and LVMI. There were no differences in other serum biomarkers, echocardiographic characteristics, or CMRI measurements among the different sST2/LVMI ratio groups (all P values > 0.05).

In this study, we constructed three models to analyze the independent effects of the sST2/LVMI ratio on the composite outcome using multivariate Cox regression analysis. The effect sizes (hazard ratios [HRs]) and 95% confidence intervals (CIs) are listed in Table 2. In the crude model, the sST2/LVMI ratio showed a positive correlation with the composite outcome (HR 1.24, 95% CI 1.03 to 1.51, P = 0.00258). In the minimally adjusted model (adjusted for sex and age), the results were similar (HR 1.25, 95% CI 1.02 to 1.53, P = 0.033), which means that for each additional 0.1-unit change in the sST2/LVMI ratio, the risk of readmission for HF increased by 25%.

Next, we analyzed the nonlinear relationship between the sST2/LVMI ratio and the composite outcome (Fig. 1). The smooth curve and the result of the Cox proportional hazards regression model with cubic spline functions showed that the relationship between the sST2/LVMI ratio and the composite outcome was positive and linear after adjusting for sex, age, body mass index, diastolic blood pressure, systolic blood pressure, and heart rate. No nonlinear relationships were observed. The Cox proportional hazard model and the two-piecewise Cox balanced hazard model were used to fit the association based on the P-value from the log likelihood ratio test (Table 3).

As shown in Table 4, only a small number of interactions were observed: age, sex, systolic blood pressure, serum uric acid, and high-density lipoprotein cholesterol (all P values for interaction < 0.05). In the present study, stronger associations were observed in patients older than 60 years (HR 3.77 [0.93, 15.26], P = 0.0380), female patients (HR 4.18 [1.08, 16.16], P = 0.014), and for systolic blood pressure ≥ 140 mmHg (HR 3.66 [0.98, 13.65], P = 0.046), serum uric acid < 416 μmol/L (HR 2.43 [1.39, 4.25], P = 0.0052), and high-density lipoprotein cholesterol ≥ 0.9 mmol/L (HR 2.16 [1.27, 3.67], P = 0.0361).

Kaplan–Meier curves estimated the composite outcome-free survival according to the sST2/LVMI ratio tertiles (Fig. 1). Patients with a high sST2/LVMI ratio (≥ 0.39), had shorter event-free survival than patients with an intermediate (between 0.39 and 0.24) or low (< 0.24) sST2/LVMI ratio (log-rank, P = 0.022). As shown in Fig. 2, there were eight, six, and two participants who reached the composite endpoint in the high, intermediate, and low groups, respectively.