Abstract
PH-CLD patients were identified from the ASPIRE registry (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019).Pectoralis muscles were automatically segmented using a deep learning model.Muscle quantity was assessed by volume and maximal cross-sectional area, indexed to patient's height.Muscle quality was assessed by mean density (Hounsfield units, HU) and intramuscular fat percentage (using established -190 to -30 HU threshold).Multivariable Cox regression adjusted for age, sex, WHO functional class, pulmonary vascular resistance (PVR) and incremental shuttle walk test distance.185 patients (124 deaths) were identified.Muscle area independently predicted mortality (HR 0.94, p = 0.040); volume was not prognostic.Muscle quality independently predicted mortality: mean density (HR 0.98, p = 0.042) and intramuscular fat percentage (HR 1.03, p = 0.015).Kaplan-Meier analysis demonstrated significantly worse survival with lower mean density (Figure 1B, log-rank p = 0.001).Deep learning-quantified pectoralis area, density and intramuscular fat percentage independently predicts mortality in PH-CLD.This is an important finding given the advent of novel antifibrotic and vasodilated therapies that require risk stratification to identify patients who would most benefit from these expensive therapies.This work adds to the increasingly recognised importance of comorbidities in PH.It also highlights the benefit of automated deep learning approaches on routine CT imaging data.Larger multicentre studies to further validate the findings are warranted, and in progress given the results of this study.