Scientific Abstracts 853 Objectives: To compare drug survival and retention rate of methotrexate (MTX) and the frst line of biotechnological drugs (b-DMARDs) in PsA patients with hypovitaminosis D and those with normal level of vitamin D. Methods: We conducted a retrospective study on 250 PsA patients (age 57,3 years ± 13,2). All patients were required to fulfll the CASPAR criteria for PsA and were tested for vitamin D levels at baseline visit. Clinical characteristics, MTX and frst line of b-DMARDs treatment duration and comorbidities information were recorded for each patient. The evaluations of drug survivals were realized by Kaplan-Meier estimate, followed by log-rank (Mentel-Cox) test for the compari- son between the groups of patients in study. Statistical signifcance was set at p ⩽ 0.05. Results: Sixty-four percent of PsA patients presented hypovitaminosis D (22,2ng/ml ± 8,8). PsA patients with hypovitaminosis D and those with normal levels were comparable for age (56,8 years ± 13 vs 58,5 years ± 12), and dis- ease activity at baseline visit (DAS 28 VES: 4 ± 0,8 vs 3,8 ± 0,8). MTX mon- otherapy survival was shorter in hypovitaminosis D group (90 ± 19 weeks vs 166,8 ± 28 weeks, p=0,041), with discontinuation risk hazard ratio = 1,4 (95% CI: 1,005 – 2,014; p=0,046). The drug survival of frst line of b-DMARDs was signifcantly shorter in patients with hypovitaminosis D (246,1 ± 40 weeks vs 302,1 ± 35 weeks; p=0,048), with discontinuation risk hazard ratio = 1,5 (95% CI: 1,1-2,4; p=0,05). Conclusion: Vitamin D seems play an important role not only in the regulation of immune system but also in the modulation on immune response induced by drugs, as MTX or b-DMARDs. The evaluation of sera levels of vitamin D at the begging of immunomodulatory therapy could have a predictive role on treatment management in PsA patients. Further studies should be useful to detect if sup- plementation of vitamin D could improve the performance of immunomodulatory drugs. Disclosure of Interests: None declared DOI: 10.1136/annrheumdis-2022-eular.5221 POS1062 HARNESSING THE POWER OF MACHINE LEARNING TO PREDICT REMISSION IN PATIENTS WITH PSORIATIC ARTHRITIS ON SECUKINUMAB: IMPLEMENTATION AND VALIDATION OF A CANDIDATE ALGORITHM ON 121 PATIENTS V. Venerito 1 , M. Fornaro 1 , F. Cacciapaglia 1 , S. Tangaro 2 , G. Lopalco 1 , F. Iannone 1 . 1 University of Bari “Aldo Moro”, Rheumatology Unit-Department of Emergency and Organ Transplantations, Bari, Italy; 2 National Institute of Nuclear Physics, Rheumatology Unit-Department of Emergen, Bari, Italy Background: Although novel therapies with biotechnological agents and small molecules may lead to the complete clearing of psoriasis in the vast majority of patients, the latter drugs only allow Psoriatic Arthritis (PsA) disease control in up to 50% of patients (1). In an increasing number of clinical scenarios, machine learning (ML) is emerging as a tool for the implementation of multi-paramet- ric decision algorithms. ML allows to handle complex non-linear relationships between patient attributes that are hard to model with traditional statistical meth- ods, merging them to output a forecast or a probability for a given outcome, enabling personalized medicine (2). Objectives: We aimed to develop a ML algorithm capable of predicting the probability of remission in PsA patients on Secukinumab to support clinicians in choosing the optimal treatment strategy. Methods: Patients with classifed PsA according to CASPAR criteria under- going Secukinumab treatment between September 2017 and September 2020 at our tertiary Centre were retrospectively observed.Either at treatment baseline and at 12-month follow up, we retrieved demographic and clinical characteristics, including Body Mass Index (BMI), disease phenotypes, Dis- ease Activity in PsA (DAPSA), Leeds Enthesitis Index (LEI) and Ankylosing Spondylitis Disease Activity Score (ASDAS, on C-Reactive Protein). After a ML variable selection method, based on an eXtreme Gradient Boosting (XGBoost) wrapper, an attribute core set with the least number of predictors was used for implementing n.3 ML algorithms, namely Logistic Regression (LR), Decision Trees (DT) and XGBoost. Each algorithm was trained and val- idated with 10-fold cross-validation. The performance of each algorithm in both phases was assessed in terms of of accuracy and area under receiver operating characteristic curve (AUROC). Results: The dataset consisted of n.121 PsA patients (62/121 female, 51.2%), with mean age (±SD) 52.9±10.1 years and mean disease duration of 5.9 ±10.4 years. Twenty-fve of them (20.7%) had axial involvement whereas 88/121 (72.7%) had polyarticular involvement. Psoriasis was present in 84/121 patients (69.4%). At baseline, mean DAPSA was 14.9 ± 9.2, mean HAQ-DI 1.1 ± 0.7, mean LEI 0.6 ± 1, mean ASDAS 2.5 ± 0.8, mean PASI 2 ± 2.9, mean BMI 28.4 ± 4.9 . Secukinumab at 300 mg dose was administered to 79/121 patients (65.3%). At 12 months DAPSA remission was achieved by 24/121 patients (19.8%). Accu- racy of LR, DT and XGBoost was of 0.70 ± 0.11, 0.81 ± 0.07 and 0.89 ± 0.05, respectively. Consistently AUROC (Figure 1 Panels ABC) were 0.63 ± 0.2, 0.79 ± 0.2 and 0.93 ± 0.1, respectively. A sample decision tree explaining XGBoost algorithm function has been provided (Figure 1 Panel D). LEI and DAPSA at baseline were shown as the most important attributes for such algorithm (Fig- ure 1 Panel E). Figure 1. Conclusion: ML can support Rheumatologists in profling those patients more likely to respond to Secukinumab. REFERENCES: [1] Scher JU, Ogdie A, Merola JF, Ritchlin C. Preventing psoriatic arthritis: focus- ing on patients with psoriasis at increased risk of transition. Nat Rev Rheu- matol. 2019 Mar;15(3):153-166. doi: 10.1038/s41584-019-0175-0. [2] Venerito V, Angelini O, Cazzato G, Lopalco G, Maiorano E, Cimmino A, Ian- none F. A convolutional neural network with transfer learning for automatic discrimination between low and high-grade synovitis: a pilot study. Intern Emerg Med. 2021 Sep;16(6):1457-1465. doi: 10.1007/s11739-020-02583-x. Epub 2021 Jan 2. Disclosure of Interests: Vincenzo Venerito Speakers bureau: Abbvie, Paid instructor for: Pfzer, Lilly, Marco Fornaro: None declared, Fabio Cacciapaglia Speakers bureau: Lilly, Abbvie, BMS. Pfzer, Paid instructor for: Lilly, Sabina Tan- garo: None declared, Giuseppe Lopalco Speakers bureau: SOBI NOVARTIS BMS ABBVIE, Paid instructor for: PFIZER, Florenzo Iannone Speakers bureau: Abbvie Pfzer UCB BMS Galapagos Novartis Lilly SOBI ROCHE, Paid instructor for: pfzer DOI: 10.1136/annrheumdis-2022-eular.5267 on February 7, 2024 by guest. Protected by copyright. http://ard.bmj.com/ Ann Rheum Dis: first published as 10.1136/annrheumdis-2022-eular.5267 on 23 May 2022. Downloaded from