OP0058 PREDICTION OF LOW DISEASE ACTIVITY IN PATIENTS WITH ANKYLOSING SPONDYLITIS TREATED WITH SECUKINUMAB IN REAL WORLD – DATA FROM THE GERMAN AQUILA STUDY

Background Secukinumab (SEC) proved to be an effective treatment for patients suffering from ankylosing spondylitis (AS) in randomized clinical trials [1]. There is only limited knowledge on prediction of low disease activity (LDA) and strategy AS under SEC routine care. Objectives Using real-world data the German non-interventional study AQUILA [2], main objectives were (1) predict LDA individual treated with through machine learning methods (2) identify most important predictors their influence using explainable artificial intelligence (XAI). Methods Data 580 used. Thirty-two demographic, parameters at baseline (BL) served as input develop models. was defined Bath index (BASDAI) ≤ 2.0 week (w) 16 (+/- 6 w). Samples divided into training (70%) validation (30%) cohorts. Ten different models applied compared. Model performance measured area receiver operating characteristic curve (AUROC) which represents probability that a randomly selected patient will have higher achieve than moderate/high activity. Additionally, sensitivity specificity model computed express proportion correctly identified who reach or don’t w16, respectively. Shapley XAI estimated importance impact each predictor based how it affected change [3]. Results The influencing BASDAI BL, followed by number pretreatments biologics, C-reactive protein (CRP), assessment spondyloarthritis international society health (ASAS-HI) height ( Figure 1 A ). AUROC best performing 0.84. Sensitivity 0.87 0.67, Applied approach showed lower BL values BASDAI, ASAS-HI biologics were, reaching w16 was. opposite case CRP body also provided visual explanations patient-individual predictions: Variables shown green color increased whereas red ones effect B Conclusion promising accuracy could reached validated. Identified such direction mostly match existing [4]. analysis can provide useful insights predictions, potentially guiding decisions future. References [1]Baeten D. et al., 2015 [2]Kiltz U. 2019 [3]Molnar C., 2022. [4]Kiltz Annals Rheumatic Diseases;79:436-437 (2020, THU0399) 1. Main results : [3] Explanation 88% Assessment (ASAS-HI), (BASDAI), mass (BMI), global (PGA), Additive exPlanations (SHAP) Acknowledgements: NIL. Disclosure Interests Asmir Vodencarevic Employee of: Novartis Pharma GmbH, Jan Brandt-Juergens Consultant Abbvie, Affibody, BMS, Gilead, Janssen, Lilly, Medac, MSD, Novartis, Pfizer, Roche, Sanofi-Aventis, UCB, Daniel Peterlik Benjamin Gmeiner Uta Kiltz AbbVie, Amgen, Biogen, Chugai, Eli GSK, Grünenthal, Hexal, Grant/research support from: UCB.