Background:The staging of interstitial lung disease (ILD) is important to monitor disease progression and for prognostication. A disease severity scale of Systemic Sclerosis (SSc)-related lung disease has long been proposed (i.e. Medsger’s severity scale). This scale was mostly developed by discussion and consensus and stage thresholds were not computed by a data-driven approach. Hidden Markov models (HMM) are methods to estimate population quantities for chronic diseases with a staged interpretation which are diagnosed by markers measured at irregular intervals.Objectives:To build a SSc-ILD specific disease severity scale with prognostic relevance via HMM modeling.Methods:A total of 358 SSc patients at risk for or with ILD were enrolled in a discovery (207 cases, Milan1) and in a validation (151 cases, Milan2, Pavia and Rome) cohort. Patients were included if satisfied the following criteria: 1) Diagnosis of SSc according to the EULAR/ACR 2013 criteria, 2) absence of anticentromere antibodies, 3) dcSSc subset or 4) other subsets with either 4a) ILD-related antibodies (Scl70, PmScl, Ku) or 4b) evidence of ILD on HRCT, 5) disease duration < 5 years at the time of the first pulmonary function test (PFT). Serial PFTs were retrieved and the time up to the last available visit -if the patient alive-, or to death due to pulmonary complications, was recorded. HMM were used to estimate the threshold of a 3-stage model (SL3SI, Scleroderma Lung 3-Stage Index) based on PFT functional values (normal/mild, moderate, severe involvement) in the discovery cohort. Survival estimates of the SL3SI model were compared to Medsger’s severity classes estimates and their predictive capability evaluated via the explained residual variation (R²) of prediction errors (the higher the better). One-hundred random replicates were generated to simulate the prediction effort in patients with different disease duration and lung severity.Results:Patients characteristics are summarized in the Table. Fifteen-years survival estimates for Mesdger’s classes in the discovery set were: normal=0.88, mild=0.86, moderate=0.84 and severe=0.71. The SL3SI was defined by the following thresholds: normal/mild, FVC and DLco >=75%; moderate FVC or DLco 74-55%; severe, FVC or DLco <55%. SL3SI 15-yrs survival estimates were: normal/mild=0.89, moderate=0.82 and severe=0.63. Prediction analysis showed a higher R²values at 15 yrs for the SL3SI compared to Medsger’s classes, providing evidence for a better predictive capability of the former (discovery: 0.31 vs 0.25; validation: 0.28 vs 0.19).Conclusion:The SL3SI, a simplified 3-stage functional model of SSc-ILD, yields better survival estimates and long-term prognostic information than Medsger’s classes. Its reproducibility and ease of use make it a useful tool for the functional and prognostic evaluation of SSc patients at risk for or with ILD.Table:VariablesDiscovery (n=207)Replication (n=151)DcSSc62 (30%)98 (64%)Age at first PFR48.6±1249.1±14.4Disease duration at first PFR1.7±1.61.3±2.4FVC90.5±18.191.1±20.2DLco70.7±19.861.3±20.1ILD on HRCT179 (86%)125 (80%)Scl70157 (76%)153 (78%)SSA63 (30%)32 (21%)n of visits38571473Follow-up time, yrs11±5.610.6±5.7Deaths27 (13%)23 (15%)Disclosure of Interests:Alessandro Santaniello: None declared, Chiara Bellocchi: None declared, Luca Bettolini: None declared, Marcello Cassavia: None declared, Gaia Montanelli: None declared, Adriana Severino: None declared, Monica Caronni: None declared, Corrado Campochiaro Speakers bureau: Novartis, Pfizer, Roche, GSK, SOBI, Enrico De Lorenzis: None declared, Gerlando Natalello: None declared, Paolo Delvino: None declared, Claudio Tirelli: None declared, Lorenzo Cavagna: None declared, Giacomo De Luca Speakers bureau: SOBI, Novartis, Celgene, Pfizer, MSD, Silvia Laura Bosello: None declared, Lorenzo Beretta Grant/research support from: Pfizer