In The American journal of cardiology ; h5-index 64.0
Unsupervised machine learning (phenomapping) has been used successfully to identify novel subgroups (phenogroups) of heart failure with preserved ejection fraction (HFpEF). However, further investigation of pathophysiological differences between HFpEF phenogroups is necessary to help determine potential treatment options. We performed speckle-tracking echocardiography and cardiopulmonary exercise testing (CPET) in 301 and 150 patients with HFpEF, respectively, as part of a prospective phenomapping study (median age 65 [25th to 75th percentile 56 to 73] years, 39% Black individuals, 65% female). Linear regression was used to compare strain and CPET parameters by phenogroup. All indicies of cardiac mechanics except for left ventricular global circumferential strain worsened in a stepwise fashion from phenogroups 1 to 3 after adjustment for demographic and clinical factors. After further adjustment for conventional echocardiographic parameters, phenogroup 3 had the worst left ventricular global longitudinal, right ventricular free wall, and left atrial booster and reservoir strain. On CPET, phenogroup 2 had the lowest exercise time and absolute peak oxygen consumption (VO2), driven primarily by obesity, whereas phenogroup 3 achieved the lowest workload, relative peak oxygen consumption (VO2), and heart rate reserve on multivariable-adjusted analyses. In conclusion, HFpEF phenogroups identified by unsupervised machine learning analysis differ in the indicies of cardiac mechanics and exercise physiology.
Dixon Debra D, Beussink-Nelson Lauren, Deo Rahul, Shah Sanjiv J
2023-Mar-06
