Dynamic prediction of pulmonary hypertension in systemic sclerosis using landmark analysis

Pulmonary hypertension (PH) is a serious complication of systemic sclerosis (SSc). We explore prediction of short-term risk for PH using serial pulmonary function tests (PFTs) and other disease features.Subjects with SSc, disease onset≥10 years prior to data retrieval, available autoantibody specificity and PFTs were included. Mixed effects modelling was used to describe change in PFTs over time. Landmarking was utilized to include serial assessments and stratified Cox proportional hazards regression analysis with landmarks as strata was used to develop the PH prediction models.We analysed 1247 SSc patients, 16.3% male, 35.8% with dcSSc. Anticentromere, anti-topoisomerase and anti-RNA polymerase antibodies were observed in 29.8%, 22.0% and 11.4% respectively and PH developed in 13.6%. Over time diffusing capacity for carbon monoxide (DLco) and carbon monoxide transfer coefficient (Kco) declined in all SSc patients (up to 1.5%/year) but demonstrated much greater annual decline (up to 4.5% and 4.8% respectively) in the 5-7 years preceding PH diagnosis. Comparison between multivariable models including either DLco, Kco or FVC/DLco ratio, demonstrated that both absolute values and change over preceding year in those measurements associate strongly with risk of PH (HR 0.93 and 0.76 for Kco and its change; HR 0.90 and 0.96 for DLco and its change; and HR 1.08 and 2.01 for FVC/DLco ratio and its change; p0.001 for all). The Kco based model had the greatest discriminating ability (Harrell's C 0.903).Our findings strongly support the importance of PFT trends over time in identifying patients at risk of PH. This article is protected by copyright. All rights reserved.