Supervivencia tras dispositivo de asistencia ventricular izquierda: evaluación de un modelo de riesgo de la base de...

Several Left Ventricular Assist Device (LVAD) risk models exist however, currently there is only one LVAD available, therefore we assessed prognostic accuracy of including data from other durable LVADs vs fully magnetic devices only in 3-year survival models.

All patients receiving primary continuous flow LVADs in The Society of Thoracic Surgeons (STS) National Intermacs Database (1/2013-4/2023) were included. Patients were stratified into 3 cohorts: Cohort 1 (fully magnetic devices only, 2018-2023, n=12,449), Cohort 2 (hydrodynamic and magnetically levitated and axial flow, 2018-2023, N=3,235), and Cohort 3 (hydrodynamic and magnetically levitated and axial flow devices, 2013-2017, n=14,019). Machine learning multivariable Oblique Random Survival Forests compared 3 models predicting 3-year survival in testing sets from the fully magnetic cohort. Model 1 included only Cohort 1 patients, Model 2 included Cohorts 1 and 2, and Model 3 included Cohorts 1,2, and 3.

Among patients with fully magnetic LVAD, 2,552 (20.5%) died within 3-years of implantation. Model discrimination, assessed by mean time dependent AUCs, was similar for all three models (0.706, 0.700, and 0.706) Calibration evaluated by time dependent delta Brier score demonstrated Model 1 had superior performance compared to Model 2 (p=0.016) and Model 3 (p=0.038), particularly after the first-year post-implant.

Models including data from older/alternative durable LVADs showed similar or inferior performance compared to a Model using fully magnetic device-only patients. These finding support current methodology including only patients receiving fully magnetic LVAD, while also providing a framework for model comparisons in future device eras.