Abstract Body:
Introduction Left Ventricular Assist Device (LVAD) implantation improves quality of life and extends survival in advanced heart failure (HF) patients. According to Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS), about 9 to 40% of the patients post-LVAD develop RHF, which significantly lowers survival and outcomes. There are several RHF risk scores for LVAD candidates, but the heterogeneity in derivation and validation limits the accuracy of RHF prediction.
Hypothesis The myocardial mRNA alternative splicing (AS) landscape will improve RHF prediction models used for risk stratification of LVAD patients.
Methods LV apical core tissue obtained during LVAD implantation at the University of Rochester Medical Center was used for bulk RNA-sequencing. We compared the transcriptome of 17 LVAD patients who developed RHF and 17 who did not develop RHF within 2 years of surgery. In addition to pair-wise gene expression analysis, we interrogated mRNA splicing of alternative exons to calculate percent spliced-in (PSI) values across the transcriptome. We utilized leave-one-out cross-validation (LOOCV) to develop AS-based logistic regression and Random Forest RHF prediction models. RT-PCR and qPCR were conducted to validate the splicing events of candidate AS events. We compared the AS risk score to a risk model using LOOCV of clinical data available for 193 patients receiving LVAD implantation over the same timeframe.
Results An L1-penalized logistic regression classifier trained on clinical features achieved only a modest discriminatory performance (AUC-ROC = 0.592). In contrast, the alternative splicing-based prediction model of RHF showed classification accuracy of 76.5% (95% CI: 58.8-89.3%) and AUC-ROC=0.834, outperforming the clinical risk model. We identified 27 alternative splicing (AS) events that were unique to the patients with RHF and have confirmed with RT-PCR and qPCR.
Conclusions Transcriptomic analysis of alternative splicing events can be utilized to predict RHF in LVAD patients. Integration of transcriptomic data with clinical predictors may facilitate development of a new and accurate RHF prediction model.