Multicenter Development and Validation of a Machine Learning Model to Predict Myocardial Recovery During LVAD Support: The UCAR Score
Recommended Citation
Kyriakopoulos CP, Taleb I, Wever-Pinzon O, Selzman C, Bonios M, Dranow E, Wever-Pinzon J, Yin M, Tseliou E, Stehlik J, Alharethi R, Kfoury A, Hanff T, Fang J, Koliopoulou A, Sideris K, Krauspe E, Nelson M, Elmer A, Singh R, Psotka M, Birks E, Slaughter M, Koenig S, Kyvernitakis A, Hoffman K, Guglin M, Kotter J, Campbell K, Silvestry S, Vidic A, Raval N, Mehra M, Cowger J, Kanwar M, Shah P, and Drakos SG. Multicenter Development and Validation of a Machine Learning Model to Predict Myocardial Recovery During LVAD Support: The UCAR Score. J Heart Lung Transplant 2023; 42(4):S106.
Document Type
Conference Proceeding
Publication Date
4-1-2023
Publication Title
J Heart Lung Transplant
Abstract
Purpose: Although significant cardiac reverse remodeling is a prerequisite for a left ventricular (LV) assist device (LVAD) patient to be considered for device weaning, multiple factors including patient goals, physician comfort, and center experience, weigh in on this complex decision. Existing predictive models defining recovery as device withdrawal, entail the above-mentioned confounders, and may under detect patients that could benefit from a targeted bridge to recovery strategy. We sought to derive and validate a predictive tool to identify patients prone to reverse remodel, independent of the complex decision to remove a durable, surgically deployed device.
Methods: Heart failure patients (N=782) requiring LVAD were enrolled at one (n=537) and five US programs (n=245). Baseline characteristics were recorded. The primary outcome was responder incidence, defined as follow-up LV ejection fraction ≥40% and LV end-diastolic diameter ≤6 cm within one year on LVAD support. Bootstrap imputation and lasso variable selection techniques were used to derive a predictive model which was then validated using our multicenter dataset. A predictive calculator was developed, and patients were classified into groups with varying potential for reverse remodeling.
Results: Patients were predominantly white (84%), male (82%), aged 56±1 years. Overall, 14.8% patients were identified as responders. Nine preoperative variables associated with reverse remodeling were included in the multivariate model achieving an optimism corrected C-statistic of 0.77 (95% CI: 0.71-0.82) (Figure).
Conclusion: The UCAR calculator is a machine learning-based multicenter and validated risk tool, implementing routine clinical data, that effectively stratifies patients into groups with varying potential for reverse remodeling. This tool can be useful in selecting patients to implement diagnostic and therapeutic protocols that can promote reverse remodeling and myocardial recovery.
Volume
42
Issue
4
First Page
S106