Echocardiographic Phenotypes Predict Long-Term Outcomes in Asymptomatic HF

Echocardiographic phenotypes significantly improved the prognostic performance on top of the ARIC HF risk score, and discriminative values were “consistently observed on top of traditionally defined cardiac structure or function abnormalities.”

Echocardiographic-based classification can identify asymptomatic patients’ long-term heart failure (HF) risk, according to research results published in JACC: Cardiovascular Imaging.

Using data from the STANISLAS Cohort ( Identifier: NCT01391442), researchers sought to verify the external validity of echocardiographic phenotyping by quantifying the phenotype with long-term incident HF and cardiovascular death risk.

The STANISLAS cohort is a single-center, familial, longitudinal, population-based cohort of those living in the Nancy region of France, established between 1993 and 1995.

Patients underwent echocardiographic examinations measuring left atrial (LA) volume, diastolic function, mitral inflow pattern, E/A ratio, e’ mean, and E/e’ mean, and left ventricular (LV) systolic deformation. Other markers, including central and peripheral blood pressures and augmentation index and inflammatory mediators associated with diastolic dysfunction in HFpEF were also assessed.

A cluster analysis was performed based on echocardiographic data to identify echocardiographic patterns through the K-means R program. In total 30 different quality measurements for different numbers of clusters were used to find the optimal number of clusters.

A toal of 827 participants (mean age, 60±5 years; 48.2% men) were included in the study. An additional 1394 participants from the Malmö Preventive Project Cohort—a population-based longitudinal cohort of inhabitants of Malmö, Sweden—who underwent echocardiogram between 2002 and 2006 and had no history of HF were also included (mean age, 67±6 years; 70% men).

Cluster analysis found 3 groups with different echocardiographic phenotypes. The largest cluster (n=334), labeled “mostly normal” had the highest e’ and E/A ratio and the highest absolute LV systolic strain. The next largest cluster included 323 patients and was labeled the “diastolic changes” phenotype, had lower e’ and higher E/e’ ratios. The final cluster (n=170)—labeled the “diastolic changes with structural remodeling” phenotype—had the highest LV mass and volumes, highest LA volume, and lowest absolute LV systolic strain. e’ was lower and E’e ratio was higher in this group.

DD Class 2016 assessment with conventional echo variables was very low across all 3 phenotypes; roughly one-half of those with the diastolic changes with structural remodeling phenotype had additional echo biomarkers.

Similar patterns of echocardiographic phenotypes were noted between K-means clustering and hierarchical clustering. Additionally, the LCM approach showed consistent phenotypes with K-means clustering.

The diastolic changes phenotype primarily included women, while the diastolic changes with structural remodeling phenotype primarily included men. Both phenotypes were associated with older age, higher BMI, and more cardiovascular risk factors compared with the mostly normal phenotype; no significant differences in these clinical features were observed.

Those in the mostly normal phenotype had more favorable levels of central and peripheral pressure, as well as vascular stiffness, compared with the diastolic changes and diastolic changes were structural remodeling phenotypes.

Among 32 circulating biomarkers, 14 were significantly different across all 3 phenotypes. These biomarkers were associated with different pathophysiological domains, such as inflammation and extracellular matrix remodeling. The diastolic phenotype had the highest levels of circulating biomarkers associated with inflammation, while biomarkers generally associated with remodeling—GDF15, PIIINP, ST-2, troponin-1, and CNP—increased from the mostly normal to diastolic and diastolic with structural remodeling phenotypes.

A decision tree identified e’, LVEDVi, and LVMi as the most relevant variables in the classification of participants in an echocardiographic profile, with a good global accuracy at 79%. The addition of clinical variables like age, sex, BMI, hypertension, diabetes, dyslipidemia, coronary artery disease, and smoking did not modify the decision algorithm.

The Malmö Preventive Project Cohort was utilized to externally validate the e’VM algorithm phenotypes. The mostly normal phenotype (n=440) had the most favorable values in terms of clinical, biological, and echocardiographic profiles. The diastolic phenotype (n=512) had the lowest proportion of men and the diastolic with structural remodeling phenotype had the highest levels of NT-proBNP and the largest LV mass and LA surface area indices.

Over a median follow-up period of 10.3 years (range, 9.8 to 11.1), 10.1% of of participants in the Malmö cohort met the primary study outcome. Compared with the mostly normal phenotype, the diastolic and diastolic with structural remodeling phenotypes were associated significantly with increased rates of the primary outcome (crude hazard ratio [HR], 2.47; 95% CI, 1.38-4.41 and crude HR, 4.67; 95% CI, 2.67-8.14).

After adjustment for ARIC HF risk score and NT-proBNP, these phenotypes remained significantly associated with the primary outcome (adjusted HR, .187 and 3.02; 95% CI, 1.04-3.37 and 1.71-5.34).

Echocardiographic phenotypes significantly improved the prognostic performance on top of the ARIC HF risk score, and discriminative values were “consistently observed on top of traditionally defined cardiac structure or function abnormalities.”

Study limitations include the observational design, a low number of participants with incident HF, and the young age of the cohort, among other limitations.

“These echocardiographic phenotypes shed new light on our understanding of asymptomatic cardiac dysfunction, and our findings may have significant clinical implications in the design of HF prevention strategies,” the researchers concluded. “A further prospective multicenter study is needed to assess the applicability of the e’VM algorithm.”

Disclosure: Several study authors declared affiliations with the pharmaceutical industry. Please see the original reference for a full list of authors’ disclosures.


Masatake K, Huttin O, Magnusson M, et al; on behalf of the STANISLAS Study. Machine learning-derived echocardiographic phenotypes predict heart failure incidence in asymptomatic individuals. JACC Cardiovasc Imaging. Published online September 15, 2021. doi: 10.1016/j.jcmg.2021.07.004