P583 Infliximab Monitoring in Crohn's Disease and Development of Neural Networks Using Clinical Variables to Predict Disease Activity and Immunogenicity

Abstract

Abstract Background Infliximab (IFX) is a biological drug that inhibits the action of tumor necrosis factor-alpha (TNFα), a pro-inflammatory cytokine enrolled in the pathogenesis of Crohn's disease (CD). Trough-level drug monitoring (TDM) has been increasingly used in clinical practice to optimize biological therapy and improve precision medicine. Therefore, this study aimed to analyze the serum level of IFX and antidrug antibodies (ADA) in active and in remission CD patients and develop prediction models via neural networks using a combination of clinical variables. Methods Seventy-five CD patients who underwent IFX therapy in the maintenance phase were included. Disease activity was defined by endoscopic and/or radiological criteria, comprising remission (CDR) and active (CDA) groups. Serum IFX levels were measured by ELISA (Promonitor ®) and rapid lateral flow test (Quantum-Blue®). ADA was measured using ELISA (Promonitor ®). For statistical analysis, a non-parametric test was used, with adopted p≤0.05. To create a prediction model, four sets of neural networks were constructed for the following output variables: remission/activity, presence or not of ADA, therapeutic level according to ELISA, and therapeutic level according to the lateral flow test. The neural networks and ROC curves were designed using the Keras package in Python software. The study was approved by the Research Ethics Committee. Results No differences were observed in the IFX level when comparing the CDA and CDR groups in both tests (p> 0.05). There was no statistical difference in IFX levels according to the use of immunosuppressants (IMS) (p>0.05). ADA levels > 5AU/mL were detected in only 11 (14.6%) patients. All patients with positive ADA had infratherapeutic IFX levels in both tests, and 72.7% were on combination therapy with IMS. Of the 4 neural networks developed, two, whose output variables were “remission/activity” and “presence or not of ADA,” showed excellent performance, with AUC ranging from 82% to 92% and 100%, respectively. Conclusion The introduction of IFX monitoring may allow more individualized and precise therapeutic management. Regarding the performance characteristics of the neural networks, an AUC >80% was evident in two of them. Through clinical and non-invasive laboratory variables, disease activity was determined. Furthermore, the finding of combinations of variables that determine the ADA level, and at the same time not having demonstrated good models for IFX serum level measurement, means that the latter continues to be necessary for clinical practice. Otherwise, ADA levels, which indicate immunogenicity, can be predicted by non-invasive clinical variables.