Cardiovascular magnetic resonance (CMR) imaging data may help cardiologists better predict whether patients are at risk for death after aortic valve replacement (AVR).
Among 799 patients with severe aortic stenosis (AS) undergoing surgical or transcatheter AVR (SAVR/TAVR), machine learning identified four CMR markers as the most predictive of mortality:
extracellular fluid volume (ECV)
late gadolinium enhancement (LGE)
left ventricular end-diastolic volume index (LVEDVi)
right ventricular ejection fraction (RVEF).
"Patient outcomes are closely associated with myocardial health at the time of AVR, with these myocardial damage markers holding major promise in optimizing the timing of AVR,” Soongu Kwak, MD, Seoul National University Hospital, South Korea, and colleagues reported in the Journal of the American College of Cardiology.
Asked by theheart.org | Medscape Cardiology whether the study offers any novel insights, Douglas Johnston, MD, Cleveland Clinic, who was not involved in the study, commented that it "does add some new understanding to imaging in patients with aortic stenosis. Machine learning allows for identification of factors which might otherwise not receive a high level of attention, in this case, extracellular fluid volume and right ventricular EF."
In an accompanying editorial, Cróchán J. O'Sullivan, MD, PhD, Bon Secours Hospital, Cork, Ireland, noted that in patients with no or minimal symptoms, those with higher levels of diffuse and replacement myocardial fibrosis on preoperative CMR had worse post-AVR survival than patients with lower levels of fibrosis.
"These observations raise the intriguing hypothesis that CMR assessment of myocardial damage may be of particular value to aid clinical decision making among patients with asymptomatic severe AS, the indication for intervention of which is a debate," he wrote.
O'Sullivan noted that current guidelines recommend AVR in asymptomatic patients with severe aortic stenosis and a left ventricular ejection fraction less than 50%, but also pointed out that this recommendation is supported only by nonrandomized data and expert consensus.
Pinak B. Shah, MD, Brigham & Women's Hospital, Boston, agreed with O'Sullivan regarding the potential applications of the findings.
"There is a lot of emerging data in both the MRI literature as well as the biomarker literature looking at potentially worse outcomes in patients with aortic stenosis who have markers of fibrosis or elevated biomarkers, so I think this is just another piece of the puzzle," he said in an interview with theheart.org | Medscape Cardiology.
"I think that the real power of this will be in trying to determine whether for patients who have severe aortic stenosis but are asymptomatic or relatively asymptomatic, we should be intervening earlier than the guidelines might otherwise suggest," he added.
Random Survival Forest
For the study, the investigators used machine learning to build a random survival forest (RSF) model using 29 variables (12 demographic or clinical, four from echocardiography, and 13 from CMR), with post-AVR death as the outcome.
"We hypothesized that RSF machine learning would provide novel insights into the predictors of death in patients with severe AS undergoing AVR, and that this data-driven approach would stratify the relative importance of myocardial damage markers and identify clinically relevant nonlinear threshold effects," they write.
The model was developed in a derivation cohort of 440 patients with severe AS scheduled for SAVR and TAVR prospectively enrolled at 10 international sites and externally validated in 359 patients from five international sites. All participants underwent CMR shortly before AVR.
There were 52 deaths in the derivation cohort over a median follow-up of 3.8 years and 51 in the validation cohort over a median of 3.3 years.
The RSF model identified mortality thresholds in both cohorts and in asymptomatic patients in multivariate models adjusted for age, sex, atrial fibrillation, and SAVR vs TAVR.
An ECV greater than 27% significant predicted increased mortality risk in the derivation (hazard ratio [HR], 2.29; P = .012) and validation (HR, 2.80; P = .002) cohorts.
Predicted mortality increased with LGR up to 2% in the derivation cohort only (HR, 2.01, P = .026).
There was a nonlinear relation between mortality and LVEDVi, with ventricles ≤55 mL/m² (HR, 2.80; P = .012 derivation only) and >80 mL/m² (HR, 2.62; P = .005 derivation; HR, 3.47; P = .007 validation) both associated with increased mortality.
Similarly, RVEF ≤50% (HR, 3.34; P = .014 validation only) and >80% (HR, 3.12; P = .034 derivation; HR, 32.5; P < .001 validation) both predicted higher mortality.
Mortality prediction was consistently higher when the four CMR variables were added to clinical risk factors, the investigators noted.
The data from this study, although informative, do not provide sufficiently strong evidence for the use of CMR to reliably identify asymptomatic patients with severe AS who might benefit from AVR, Johnston cautioned.
"The difficulty in applying such knowledge to patients, in part, is a result of the design of the study," he said. "The authors include patients with coronary artery disease. While this does represent a real world population, many patients with aortic stenosis have concomitant coronary disease. It is then difficult to tease out whether the differences in mortality relate to the timing of progression of AS, as the authors suggest, or to the underlying disease of the heart muscle, as might be reflected by CAD or other concomitant disease-longstanding hypertension, etc."
O'Sullivan noted that two prospective multicenter trials are currently evaluating the hypothesis that early intervention with AVR will offer superior outcomes to observation alone in patients with severe aortic stenosis.
In the 1000-patient EVOLVED (Early Valve Replacement Guided by Biomarkers of LV Decompensation in Asymptomatic Patients With Severe AS) trial, patients with mid-wall fibrosis are randomized to receive either early surgical intervention or routine care and those with no mid-wall fibrosis are randomized to routine care with or without study follow-up.
In the EARLY-TAVR (Evaluation of Transcatheter Aortic Valve Replacement Compared to Surveillance for Patients With Asymptomatic Severe Aortic Stenosis) trial, patients are randomized to early TAVR or clinical surveillance, independent of CMR findings.
The study was funded by grants from the Republic of Korea Ministry of Science and ICT and Ministry of Health & Welfare. All authors reported having no conflicts of interest relevant to the study. O'Sullivan, Johnston, and Shah reported no relevant conflicts of interest.
Neil Osterweil, an award-winning medical journalist, is a long-standing and frequent contributor to Medscape.
Medscape Medical News © 2021
Cite this: AI Hones in on CMR Markers of Mortality in Aortic Stenosis - Medscape - Aug 11, 2021.