ObjectivesAround 30% of patients with rheumatoid arthritis (RA) do not respond to tumour necrosis factor inhibitors (TNFi). We aimed to predict patient response to TNFi using machine learning on simple clinical and biological data.MethodsWe used data from the RA ESPOIR cohort to train our models. The endpoints were the EULAR response and the change in Disease Activity Score (DAS28). We compared the performances of multiple models (linear regression, random forest, XGBoost and CatBoost) on the training set and cross-validated them using the area under the receiver operating characteristic curve (AUROC) or the mean squared error. The best model was then evaluated on a replication cohort (ABIRISK).ResultsWe included 161 patients from ESPOIR and 118 patients from ABIRISK. The key selected features were DAS28, lymphocytes, ALT (aspartate aminotransferase), neutrophils, age, weight, and smoking status. When predicting EULAR response, CatBoost achieved the best performances of the four tested models. It reached an AUROC of 0.72 (0.68–0.73) on the train set (ESPOIR). Better results were obtained on the train set when etanercept and monoclonal antibodies were analysed separately. On the test set (ABIRISK), these models respectively achieved on AUROC of 0.70 (0.57–0.82) and 0.71 (0.55–0.86). Two decision thresholds were tested. The first prioritised a high confidence in identifying responders and yielded a confidence up to 90% for predicting response. The second prioritised a high confidence in identifying inadequate responders and yielded a confidence up to 70% for predicting non-response. The change in DAS28 was predicted with an average error of 1.1 DAS28 points.ConclusionThe machine learning models developed allowed predicting patient response to TNFi exclusively using data available in clinical routine.

Objectives Around 30% of patients with rheumatoid arthritis (RA) have an inadequate response to methotrexate (MTX). We aimed to use routine clinical and biological data to build machine learning models predicting EULAR inadequate response to MTX and to identify simple predictive biomarkers. Methods Models were trained on RA patients fulfilling the 2010 ACR/EULAR criteria from the ESPOIR and Leiden EAC cohorts to predict the EULAR response at 9 months (+/- 6 months). Several models were compared on the training set using the AUROC. The best model was evaluated on an external validation cohort (tREACH). Model’s predictions were explained using Shapley values to extract a biomarker of inadequate response. Results We included 493 therapeutic sequences from ESPOIR, 239 from EAC, and 138 from tREACH. The model selected DAS28, Lymphocytes, Creatininemia, Leucocytes, AST, ALT, swollen joint count, and corticosteroid co-treatment as predictors. The model reached an AUROC of 0.72 (95% CI : [0.63; 0.80]) on the external validation set, where 70% of patients were responders to MTX. Patients predicted as inadequate responders had only 38% (95% CI : [20%; 58%]) chance to respond and using the algorithm to decide to initiate MTX would decrease inadequate-response rate from 30% to 23% (95% CI: [17%; 29%]). A biomarker was identified in patients with moderate or high activity (DAS28 > 3.2): patients with a lymphocyte count superior to 2000 cells/mm3 are significantly less likely to respond. Conclusion Our study highlights the usefulness of machine learning in unveiling subgroups of inadequate responders to MTX to guide new therapeutic strategies. Further work is needed to validate this approach.

BackgroundThirteen drugs with different mechanisms of action may be considered to treat patients with methotrexate (MTX) inadequate response in rheumatoid arthritis (RA). TNF inhibitors (TNFi) are frequently the first choice in this situation. Unfortunately, 30% to 40% of RA patients do not respond to TNFi, resulting in a delay for beginning the appropriate targeted DMARD (tDMARD). Predicting the patient response to TNFi before prescribing the treatment is therefore a major goal and could help physicians to prescribe a tDMARD suited to the patient.ObjectivesWe aimed to build machine learning models based on simple clinical and biological data to predict patient response to TNFi.MethodsWe used data from the ESPOIR early arthritis cohort (1) to train the models, and the ABIRISK cohort (2) to validate the results. We included patients that fulfilled the EULAR/ACR 2010 criteria and that were treated with a TNFi. The models take as inputs patient’s characteristics at treatment initiation and predicts the therapeutic response, defined as the EULAR response 12 months (+/- 6 months) after treatment initiation. We compared the performances of four models (Linear Regression, Random Forest, XGBoost, and Catboost) on the training set by cross-validated them using the Area Under the ROC Curve (AUCROC). The best model was then evaluated on the validation dataset (ABIRISK cohort). We conducted the methodology both on all TNFi together and on etanercept and monoclonal anti-TNF antibodies separated. We analyzed how clinical and biological variables impacted response to provide explainability of the prediction.ResultsWe included 164 patients from the ESPOIR cohort and 118 patients from the ABIRISK cohort. Better results were obtained when etanercept and monoclonal anti-TNF antibodies were analyzed separately.These models predict a probability for a patient to respond to TNFi. This probability is compared to a decision threshold to obtain the binary outcome. Two decision thresholds were tested. The first prioritizes a high confidence when identifying responders (Strategy 1) while the second prioritizes a high confidence when identifying non-responders (Strategy 2). The model’s results are presented in Table 1.Table 1.Sensitivity, Specificity, PPV and NPV are computed on the ABIRISK validation cohort for each strategyDrugAUC(ESPOIR)AUC(ABIRA)STRATEGY 1 (high confidence in response)STRATEGY 2 (high confidence in non-response)SensitivitYSpecificityPPVNPVSensitivitySpecificityPPVNPVOverall TNFi0.720.6518%91%76%42%90%30%67%67%(0.68-0.73)(0.54 - 0.75)(10%-27%)(82%-98%)(54%-95%)(32%-51%)(83%-96%)(18%-44%)(58%-76%(45%-86%Etanercept0.740.7060%73%78%53%95%15%64%67%(0.68-0.75)(0.57- 0.82)(44%-74%)(55%-89%)(63%-92%)(36%-69%)(88%-100%)(4%-30%)(52%-76%)(20%-100%)Monoclonal anti-TNF antibodies0.740.7137%95%92%50%90%40%69%73%(0.69-0.77)(0.55-0.86)(20%-55%)(83%-100%)(73%-100%)(35%-66%)(78%-100%)(19%-62%)(54%-84%)(44%-100%)Using SHAP, we were able to analyze how each variable impacted the predictions. In particular, a DAS28 around 5 had the highest positive impact on response. Higher and lower values of DAS28 had either less impact or even negative impact on the patient response to TNFi treatment (Figure 1). This allows to identify non-linear relations between variables and patient response.ConclusionThe machine learning models developed in this study can predict RA patients’ response to TNFi using exclusively data available in clinical routine. These models also allow to analyze how these variables are used to predict response. Along with similar models for other tDMARDs, such algorithms could lead to a personalized therapeutic strategy.References[1]Combe B, Benessiano J, Berenbaum F, Cantagrel A, Daurès J-P, Dougados M, et al. The ESPOIR cohort: A ten-year follow-up of early arthritis in France.[2]Anon. ABIRISK Anti-Biopharmaceutical Immunization: Prediction and Analysis of Clinical Relevance to Minimize the RISK. 2019.AcknowledgementsFor the first 5 years of the ESPOIR cohort, an unrestricted grant from Merck Sharp and Dohme (MSD) was allocated. Two additional grants from INSERM were obtained to support part of the biological database. The French Society of Rheumatology, Pfizer, AbbVie, Lilly, and more recently Fresenius and Biogen also supported the ESPOIR cohort study. We also wish to thank Nathalie Rincheval (Montpellier) who did expert monitoring and data management and all the investigators who recruited and followed the patients (F.Berenbaum, Paris-Saint Antoine, MC.Boissier, Paris-Bobigny, A.Cantagrel, Toulouse, B.Combe, Montpellier, M.Dougados, Paris-Cochin, P.Fardellone et P.Boumier Amiens, B.Fautrel, Paris-La Pitié, RM. Flipo, Lille, Ph. Goupille, Tours, F. Liote, Paris-Lariboisière, O.Vittecoq, Rouen, X.Mariette, Paris Bicetre, P.Dieude, Paris Bichat, A.Saraux, Brest, T.Schaeverbeke, Bordeaux, J.Sibilia, Strasbourg) as well as S.Martin (Paris Bichat) who did all the central dosages of CRP, IgA and IgM rheumatoid.SB was supported by FHU CARE.Disclosure of InterestsNone declared.

No abstract

No abstract