Integrated trajectories of systolic and diastolic function differentially associate with risk for heart failure with preserved and reduced ejection fraction and proteomic profiles

Aims To jointly model longitudinal measures of left ventricular ejection fraction (LVEF) and E/A ratio in late‐life, and to assess whether predicted trajectory membership is associated with heart failure risk. Methods and results Using a Bayesian non‐parametric trajectory approach, trajectories were modelled among 747 Jackson Heart Study participants who underwent ≥2 echocardiograms in 2000–2004 (age 65 ± 5 years), 2011–2013 (75 ± 5), and 2018–2019 (81 ± 5). Using the resulting model, we predicted trajectory membership for 4419 distinct Atherosclerosis Risk in Communities (ARIC) study participants based on single time‐point measures of LVEF and E/A ratio (age 75 ± 5 years; ‘testing cohort’). Multivariable Cox models assessed the relationship between predicted trajectory and incident heart failure with preserved (HFpEF) and reduced ejection fraction (HFrEF). We evaluated associations of 4877 plasma proteins (SOMAscan) with predicted trajectory and performed Mendelian randomization to assess causal effects on LVEF and volume. Six trajectories were identified: pink (prevalence 50%) and light green (17%) – increasing LVEF, decreasing E/A ratio with age; red (22%) – no increase in LVEF; dark green (4%) – declining LVEF; orange (2%) – steeply declining LVEF, rising E/A ratio; and blue (4%) – rising E/A ratio despite increasing LVEF. In the testing cohort, red and dark green associated with HFrEF alone, blue with HFpEF alone, and orange with both compared to pink. Trajectory membership provided incremental value in predicting heart failure and HFpEF. Mendelian randomization identified potential causal effects of 13 trajectory‐associated proteins on LVEF and volume. Conclusions Bayesian non‐parametric modelling identifies cardiac function trajectories differentially associated with HFpEF and HFrEF and holds promise to improve risk prediction and enable therapeutic target discovery.