death. These parameters had been defined by expert nominal group technique. Methodologically, two main limitations had to be addressed: 1 the problem of missing data and 2 the low number of events, which prohibits the simultaneous exploration of a large set of predictors in a regression model. We addressed these issues by (a) imputing multiple predictors based on different algorithms (multiple imputation), and (b) using least absolute shrinkage and selection operator (LASSO) regression for 42

more »

... inical parameters. Results: Of 1451 patients who met the inclusion criteria, 706 had complete data available on all parameters for disease worsening. Of 706 patients originally evaluated, 228 (32.3%) had disease progression, most of which was either a decrease in FVC (103 patients, 14.6%) or death (92 patients, 13.0%) within the observation period (12±3 months). Of the 42 clinical parameters introduced into the model as outcome predictors, 8 remained in the final regression model, which was chosen by the Bayesian information criterion (table 1) . Bootstrap with 10 000 repetitions successfully validated the model. Abstract FRI0456 -Table 1. Predictive factors in the final LASSO regression model Factor p-value Odds ratio Scientific Abstracts on 1 May 2019 by guest. Protected by copyright.