Parsimonious echocardiography-based model for predicting long-term all-cause mortality in heart failure with preserved ejection fraction.
The Heart Failure Association Pre-test Assessment, Echocardiography and Natriuretic Peptide, Functional testing, Final aetiology (HFA-PEFF) score and the Heavy, Hypertensive, Atrial Fibrillation, Pulmonary Hypertension, Elder, Filling Pressure (H2FPEF) score facilitate heart failure with preserved ejection fraction (HFpEF) diagnosis but may not adequately stratify mortality risk. We developed and internally validated a parsimonious echocardiography-based model for long-term all-cause mortality in HFpEF and compared its prognostic discrimination with HFA-PEFF and H2FPEF.
In a real-world HFpEF echocardiography database linked to territory-wide electronic health records, 792 adults with HFpEF (left ventricular ejection fraction ≥50%) diagnosed between 2010 and 2016 were randomly split in a prespecified 70:30 ratio into training (n=554) and validation (n=238) cohorts. The primary endpoint was all-cause mortality. A parsimonious model was derived using least absolute shrinkage and selection operator (LASSO)-penalised Cox regression and refitted as a multivariable Cox model.
The final nomogram included age, left ventricular posterior wall thickness at end-systole, mitral E velocity, E/e' ratio and pulmonary artery systolic pressure. During median follow-up of 5.17 years (IQR 2.26-9.14) in the training cohort and 5.75 years (IQR 2.17-9.25) in the validation cohort, 393/554 (70.9%) and 165/238 (69.3%) deaths occurred, respectively. In the validation cohort, the nomogram showed better discrimination than HFA-PEFF and H2FPEF, with 1/3/5-year area under the curves of 0.658/0.706/0.713 versus 0.507/0.561/0.642 and 0.516/0.533/0.607, respectively. Calibration was acceptable at 1 and 3 years but weaker at 5 years, and risk-score tertiles separated survival in both cohorts (log-rank p<0.001).
A five-variable echocardiography-based nomogram showed better discrimination for long-term mortality prediction than the repurposed diagnostic scores evaluated in this cohort. External validation is needed before clinical implementation.
In a real-world HFpEF echocardiography database linked to territory-wide electronic health records, 792 adults with HFpEF (left ventricular ejection fraction ≥50%) diagnosed between 2010 and 2016 were randomly split in a prespecified 70:30 ratio into training (n=554) and validation (n=238) cohorts. The primary endpoint was all-cause mortality. A parsimonious model was derived using least absolute shrinkage and selection operator (LASSO)-penalised Cox regression and refitted as a multivariable Cox model.
The final nomogram included age, left ventricular posterior wall thickness at end-systole, mitral E velocity, E/e' ratio and pulmonary artery systolic pressure. During median follow-up of 5.17 years (IQR 2.26-9.14) in the training cohort and 5.75 years (IQR 2.17-9.25) in the validation cohort, 393/554 (70.9%) and 165/238 (69.3%) deaths occurred, respectively. In the validation cohort, the nomogram showed better discrimination than HFA-PEFF and H2FPEF, with 1/3/5-year area under the curves of 0.658/0.706/0.713 versus 0.507/0.561/0.642 and 0.516/0.533/0.607, respectively. Calibration was acceptable at 1 and 3 years but weaker at 5 years, and risk-score tertiles separated survival in both cohorts (log-rank p<0.001).
A five-variable echocardiography-based nomogram showed better discrimination for long-term mortality prediction than the repurposed diagnostic scores evaluated in this cohort. External validation is needed before clinical implementation.