A Novel Model Incorporating Pectoralis Muscle Measures to Predict Mortality after Ventricular Assist Device Implantation

Abstract

Purpose We have previously shown that pectoralis muscle mass and tissue attenuation obtained on preoperative CT scans were powerful predictors of mortality after LVAD implantation using single center data. This analysis confirms those findings in a separate LVAD dataset and presents a novel prediction model developed across centers. Methods Patients with chest CTs performed ≤ 3 months before LVAD implantation at University of Minnesota (n=143) and Houston Methodist Hospital (n=133) were identified from larger cohorts (University of Minnesota n=353, Houston Methodist =278). Unilateral pectoralis muscle mass indexed to body surface area (PMI) and attenuation (approximated by mean Hounsfield units; PHUm) were measured on preoperative chest CT scans. To develop the Pectoralis Muscle Risk Score (PMRS), we implemented a cross-validated model selection using Cox's proportional hazards regression model with censoring at transplantation. The final model included PHUm, sex, albumin, year of implant, body mass index, bridge to transplant, INTERMACS profile, and presence or absence of contrast. Receiver-operating characteristic (ROC) curves for 30, 90, and 365 day survival were generated for the model and compared to the HeartMate II Risk Score (HMRS). Results The Pectoralis Risk Score outperformed HMRS at each of the pre-specified time-points. The AUC for PMRS for predicting survival at 30, 90, and 365 days was 78%, 76%, and 76%, respectively. The HMRS had AUC of 63%, 67%, and 63% (Figure). Conclusion The Pectoralis Muscle Risk Score had favorable discrimination in this multicenter dataset. These skeletal muscle measures, which are associated with heart failure stage and frailty, appear to add important information to pre-operative risk assessment.