SAT0039 ADAPTIVE DEEP LEARNING FOR THE PREDICTION OF INDIVIDUAL DISEASE ACTIVITY IN PATIENTS WITH RHEUMATOID ARTHRITIS. (2nd June 2020)

• Record Type: Journal Article
• Authors: Hügle, M.; Kalweit, G.; Walker, U.; Finckh, A.; Muller, R.; Scherer, A.; Boedecker, J.; Hügle, T.
• Is Part Of: Annals of the rheumatic diseases. Volume 79(2020)Supplement 1
• Page Start: 950
• Page End: 950
• Publication Date: 2020-06-02
• Subjects: Rheumatism -- Periodicals; 616.723005
• DOI: 10.1136/annrheumdis-2020-eular.2712
• Languages: English
• ISSNs: 0003-4967

Abstract

Abstract : Background: Rheumatoid arthritis (RA) lacks reliable biomarkers that predict disease evolution on an individual basis, potentially leading to over- and undertreatment. Deep neural networks learn from former experiences on a large scale and can be used to predict future events as a potential tool for personalized clinical assistance. Objectives: To investigate deep learning for the prediction of individual disease activity in RA. Methods: Demographic and disease characteristics from over 9500 patients with 65.000 visits from the Swiss Quality Management (SCQM) database were used to train and evaluate an adaptive recurrent neural network (AdaptiveNet). Patient and disease characteristics along with clinical and patient reported outcomes, laboratory values and medication were used as input features. DAS28-BSR was used to predict active disease and future numeric individual disease activity by classification and regression, respectively. Results: AdaptiveNet predicted active disease defined as DAS28-BSR>2.6 at the next visit, with an overall accuracy of 75.6% and a sensitivity and specificity of 84.2% and 61.5%, respectively. Apart from DAS28-BSR, the most influential characteristics to predict disease activity were joint pain, disease duration, age and medication. Longer disease duration, age >50 or antibody positivity marginally improved prediction performance. Regression allowed forecasting individual DAS28-BSR values with a mean squared error of 0.9. Conclusion: Deep neural networks have the capacity to predict individual disease outcome in RA. Low specificity remains challenging and might benefit from alternative input data or outcome targets. References: [1] Hügle M, Kalweit G, Hügle T, Boedecker J. A Dynamic Deep Neural Network For Multimodal Clinical Data Analysis. Be Publ Stud Comput Intell Springer Verl. 2020. Disclosure of Interests: Maria Hügle Paid instructor for: Lilly, Gabriel Kalweit: None declared, Ulrich Walker Grant/research support from: Ulrich Walker has received an unrestricted research grant from Abbvie, Consultant of: Ulrich Walker has act as a consultant for Abbvie, Actelion, Boehringer Ingelheim, Bristol-Myers Squibb, Celgene, MSD, Novartis, Pfizer, Phadia, Roche, Sandoz, Sanofi, and ThermoFisher, Paid instructor for: Abbvie, Novartis, and Roche, Speakers bureau: Abbvie, Actelion, Bristol-Myers Squibb, Celgene, MSD, Novartis, Pfizer, Phadia, Roche, Sandoz, and ThermoFisher, Axel Finckh Grant/research support from: Pfizer: Unrestricted research grant, Eli-Lilly: Unrestricted research grant, Consultant of: Sanofi, AB2BIO, Abbvie, Pfizer, MSD, Speakers bureau: Sanofi, Pfizer, Roche, Thermo Fisher Scientific, Rudiger Muller Consultant of: AbbVie, Nordic, Sandoz, Almut Scherer: None declared, Joschka Boedecker: None declared, Thomas Hügle Grant/research support from: Abbvie, Novartis, Consultant of: Abbvie, Pfizer, Novartis, Roche, Lilly, BMS