Masters candidate in Biostatistics, Meredith Hyun, will present:
“Postoperative Patient Outcomes Following Left Ventricular Assist Device Implantation: An Application Format for the Minnesota Pectoralis Risk Score Predicting Patient Mortality, and A Cross-Validated Logistic Regression Model for Predicting Early Postoperative Severe Right Heart Failure”
Plan B Adviser: Thomas Murray
Abstract: Cardiac implantation of a Left Ventricular Assist Device (LVAD) is one treatment option for patients with advanced heart failure. Though known to improve survival and quality of life for many, LVAD implantations have also been known to lead to worsening heart conditions and higher mortality rates in others. To forecast likely postoperative mortality outcomes, previous work has formulated the Minnesota Pectoralis Risk Score (MPRS). The model used preoperative patient characteristics such as chest muscle measures (mass and tissue attenuation), race, and BMI as predictors. For this project, the first aim was to embed the MPRS into an accessible format. The Shiny package in R was used to create a web application allowing the input of individual characteristics to output corresponding information about post-LVAD mortality risks. The second segment of the project constructed two logistic regression models for predicting the following outcomes after LVAD implantation: early severe right heart failure, and early Right Ventricular Assist Device (RVAD) implantation. Predictor combinations for each were selected for using two-fold cross-validation after completion of linearity checks, consideration for interactions, and pre-screening. The final severe RHF model included albumin, pectoralis muscle tissue attenuation, creatinine, pulmonary vascular resistance, INTERMACS profile, and gender as variables. Its average out-of-sample AUC was 75%, with in-sample AUC of 77% (95% CI: 68-87%). Variables for the final early postoperative RVAD model differed only in that implantable cardiac monitor presence was included, while gender was not. Average out-of-sample AUC for the RVAD model was 76%; in-sample was 85% (95% CI: 72-97%). Prediction of need for RVAD after LVAD may be particularly useful for the identification of candidates for Biventricular Assist Device (BiVAD) implantation.
Refreshments will be served prior to the presentation.